Astronomy & Astrophysics manuscript no. gd358 January 23, 2003 (DOI: will be inserted by hand later)

The Everchanging Pulsating White Dwarf GD358

S.O. Kepler1, R. Edward Nather2, Don E. Winget2, Atsuko Nitta3, S. J. Kleinman3, Travis Metcalfe2,4, Kazuhiro Sekiguchi5, Jiang Xiaojun6, Denis Sullivan7, Tiri Sullivan7, Rimvydas Janulis8, Edmund Meistas8, Romualdas Kalytis8, Jurek Krzesinski9, Waldemar OgÃloza9, Staszek Zola10, Darragh O’Donoghue11, Encarni Romero-Colmenero11, Peter Martinez11, Stefan Dreizler12, Jochen Deetjen12, Thorsten Nagel12, Sonja L. Schuh12, Gerard Vauclair13, Fu Jian Ning13, Michel Chevreton14, Jan-Erik Solheim15, Jose M. Gonzalez Perez15, Frank Johannessen15, Antonio Kanaan16, Jos´eEduardo Costa1, Alex Fabiano Murillo Costa1, Matt A. Wood17, Nicole Silvestri17, T.J. Ahrens17, Aaron Kyle Jones18,∗, Ansley E. Collins19,∗, Martha Boyer20,∗, J. S. Shaw21, Anjum Mukadam2, Eric W. Klumpe22, Jesse Larrison22, Steve Kawaler23, Reed Riddle23, Ana Ulla24, and Paul Bradley25

1 Instituto de F´ısicada UFRGS, Porto Alegre, RS - Brazil e-mail: [email protected] 2 Department of Astronomy & McDonald Observatory, University of Texas, Austin, TX 78712, USA 3 Sloan Digital Sky Survey, Apache Pt. Observatory, P.O. Box 59, Sunspot, NM 88349, USA 4 Harvard-Smithsonian Center for Astrophysics, 60 Garden Street, Cambridge, MA 02138 USA e-mail: [email protected] 5 Subaru National Astronomical Observatory of Japan e-mail: [email protected] 6 Beijing Astronomical Observatory, Academy of Sciences, Beijing 100080, China e-mail: [email protected] 7 University of Victoria, Wellington, New Zealand 8 Institute of Theoretical Physics and Astronomy, Gostauto 12, Vilnius 2600, Lithuania 9 Mt. Suhora Observatory, Cracow Pedagogical University, Ul. Podchorazych 2, 30-084 Cracow, Poland 10 Jagiellonian University, Krakow, Poland e-mail: [email protected] 11 South African Astronomical Observatory 12 Universitat T¨ubingen,Germany 13 Universit´ePaul Sabatier, Observatoire Midi-Pyr´en´ees,CNRS/UMR5572, 14 av. E. Belin, 31400 Toulouse, France 14 Observatoire de Paris-Meudon, DAEC, 92195 Meudon, France e-mail: [email protected] 15 Institutt for fysikk, 9037 Tromso, Norway 16 Departamento de F´ısica, Universidade Federal de Santa Catarina, CP 476, CEP 88040-900, Florian´opolis, Brazil, e-mail: [email protected] 17 Dept. of Physics and Space Sciences & The SARA Observatory, Florida Institute of Technology, Melbourne, FL 32901 ? 18 University of Florida, 202 Nuclear Sciences Center Gainesville, FL 32611-8300 19 Johnson Space Center, 2101 NASA Road 1, Mail Code GT2, Houston, TX 77058, USA 20 University of Minnesota, Department of Physics & Astronomy, 116 Church St. S.E., Minneapolis, MN 55455 21 University of Georgia at Athens, Department of Physics and Astronomy, Athens, GA 30602-2451, USA 22 Middle Tennessee State University, Department of Physics and Astronomy Murfreesboro, TN 37132, USA 23 Department of Physics and Astronomy, Iowa State University, Ames, IA 50011, USA 24 Universidade de Vigo, Depto. de Fisica Aplicada, Facultade de Ciencias, Campus Marcosende-Lagoas, 36200 Vigo (Pontevedra), Spain e-mail: [email protected] 25 Los Alamos National Laboratory, X-2, MS T-085 Los Alamos, NM 87545, USA

Received 6 Dec 2002 / Accepted 22 Jan 2003

Abstract. We report 323 hours of nearly uninterrupted time series photometric observations of the DBV star GD 358 acquired with the Whole Earth Telescope (WET) during May 23rd to June 8th, 2000. We acquired more than 232 000 independent measurements. We also report on 48 hours of time-series photometric observations in Aug 1996. We detected the non-radial g-modes consistent with degree ` = 1 and radial order 8 to 20 and their linear combinations up to 6th order. We also detect, for the first time, a high amplitude ` = 2 mode, with a period of 796 s. In the 2000 WET data, the largest amplitude modes are similar to those detected with the WET observations of 1990 and 1994, but the highest combination order previously detected was 4th order. At one point 2 Kepler et al.: The Everchanging Pulsating White Dwarf GD358

during the 1996 observations, most of the pulsation energy was transferred into the radial order k = 8 mode, which displayed a sinusoidal pulse shape in spite of the large amplitude. The multiplet structure of the individual modes changes from year to year, and during the 2000 observations only the k = 9 mode displays clear normal triplet structure. Even though the pulsation amplitudes change on timescales of days and years, the eigenfrequencies remain essentially the same, showing the stellar structure is not changing on any dynamical timescale.

Key words. (Stars:) white dwarfs, Stars: variables: general, Stars: oscillations, Stars: individual: GD 358, Stars: evo- lution

1. Introduction GD 358, also called V777 Herculis, is the prototype of the DBV class of white dwarf pulsators. It was the first pulsating star detected based on a theoretical prediction (Winget et al. 1985), and is the pulsating star with the largest number of periodicities detected after the Sun. Detecting as many modes as possible is important, as each periodicity detected yields an independent constraint on the star’s structure. The study of pulsating white dwarf stars has allowed us to measure the stellar mass and composition layers, to probe the physics at high densities, including crystallization, and has provided a chronometer to measure the age of the oldest stars and consequently, the age of the Galaxy. Robinson, Kepler & Nather (1982) and Kepler (1984) demonstrated that the variable white dwarf stars pulsate in non-radial gravity modes. Beauchamp et al. (1999) studied the spectra of the pulsating DBs to determine their instability strip at 22 400 ≤ Teff ≤ 27 800 K, and found Teff = 24 900 K, log g = 7.91 for the brightest DBV, GD 358 (V=13.85), assuming no photospheric H, as confirmed by Provencal et al. (2000). Provencal et al. studied the HST and EUVE spectra, deriving Teff = 27 000 ± 1 000 K, finding traces of carbon in the atmosphere [log(C/He) = −5.9 ± 0.3] and a broadening corresponding to v sin i = 60±6 km/s. They also detected Lyα that is probably interstellar. Althaus & Benvenuto (1997) demonstrated that the Canuto, Goldman, & Mazzitelli (1996, hereafter CGM) self consistent theory of turbulent convection is consistent with the Teff ' 27 000 K determination, as GD 358 defines the blue edge of the DBV instability strip. Shipman et al. (2002) extended the blue edge of the DBV instability strip by finding that the even hotter star PG0112+104 is a pulsator. Winget et al. (1994) reported on the analysis of 154 hours of nearly continuous time series photometry on GD 358, obtained during the Whole Earth Telescope (WET) run of May 1990. The Fourier temporal spectrum of the light curve is dominated by periodicities in the range 1000 – 2400 µHz, with more than 180 significant peaks. They identify all of the triplet frequencies as having degree ` = 1 and, from the details of their triplet (k) spacings, from which Bradley & Winget (1994) derived the total stellar mass as 0.61 ± 0.03 M¯, the mass of the outer helium envelope as −6 2.0 ± 1.0 × 10 M∗, the luminosity as 0.050 ± 0.012 L¯ and, deriving a temperature and bolometric correction, the distance as 42 ± 3 pc. Winget et al. (1994) found changes in the m spacings among the triplet modes, and by assuming the rotational splitting coefficient C`,k(r) depends only on radial overtone k and the rotation angular velocity Ω(r), interpret the observed spacing as strong evidence of radial differential rotation, with the outer envelope rotating some 1.8 times faster than the core. However, Kawaler, Sekii, & Gough (1999) find that the core rotates faster than the envelope when they perform rotational splitting inversions of the observational data. The apparently contradictory result is due to the presence of mode trapping and the behavior of the rotational splitting kernel in the core of the model. Winget et al. also found significant power at the sums and differences of the dominant frequencies, indicating that non–linear processes are significant, but with a richness and complexity that rules out resonant mode coupling as a major cause. We show that in the WET data set reported here (acquired in 2000), only 12 of the periodicities can be identified as independent g-mode pulsations, probably all different radial overtones (k) with same spherical degree ` = 1, plus the azimuthal m components for k = 8 and 9. The high amplitude with a period of 796 s is identified as an ` = 2 mode; it was not present in the previous data sets. Most, if not all, of the remaining periodicities are linear combination peaks of these pulsations. Considering there are many more observed combination frequencies than available eigenmodes, we interpret the linear combination peaks as caused by non-linear effects, not real pulsations. This interpretation is consistent with the proposal by Brickhill (1992) and Wu (2001) that the combination frequencies appear by the non-linear response of the medium. Recently, van Kerkwijk et al. (2000) and Clemens et al. (2000) show that most linear combination peaks for the DAV G29-38 do not show any velocity variations, while the eigenmodes do. However, Thompson et al. (2003) argue that some combination peaks do show velocity variations. As a clear demonstration of the power of asteroseismology, Metcalfe, Winget, & Charbonneau (2001) and Metcalfe, Salaris, & Winget (2002) used GD 358 observed periods from Winget et al. (1994) and a genetic algorithm to search for the optimum theoretical model with static diffusion envelopes, and constrained the 12C(α, γ)16O cross section, a crucial

Send offprint requests to: S.O. Kepler, e-mail: [email protected] ? Southeastern Association for Research in Astronomy (SARA) NSF-REU Student. Kepler et al.: The Everchanging Pulsating White Dwarf GD358 3 parameter for many fields in astrophysics and difficult to constrain in terrestrial laboratories. Montgomery, Metcalfe, & Winget (2001) also used the observed pulsations to constrain the diffusion of 3He in white dwarf stars. They show 4 3 their best model¡ for¢ GD 358 has O/C/ He/ He structure, Teff = 22 300 ± 500¡ K, M¢ ∗ = 0.630¡ ± 0.015 M¯,¢ a thick He 4 4 +0.18 layer, log M He /M∗ = (−2.79 ± 0.06), distinct from the thin layer, log M He /M¡ ∗ =¢ −5.70 ± −0.30 , proposed 3 by Bradley & Winget (1994). Montgomery, Metcalfe, & Winget’s model had log M He /M∗ = (−7.49 ± 0.12), but Wolff et al. (2002) did not detect any 3He in the spectra of all the DBs they observed. On the other hand, Dehner & Kawaler (1995), Brassard & Fontaine (2002), and Fontaine & Brassard (2002) show that a thin helium envelope is consistent with the evolutionary models starting at PG1159 models and ending as DQs, as diffusion is still ongoing around 25 000K and lower temperatures. Therefore there could be two transition zones in the envelope, one between the He envelope and the He/C/O layer, where diffusion is still separating the elements, and another transition between this layer and the C/O core. Gautschy & Althaus (2002) calculated nonadiabatic pulsation properties of DB pulsators using evolutionary models including the CGM full-spectrum turbulence theory of convection and time-dependent element diffusion. They show that up to 45 dipole modes should be excited, with periods between 335s and 2600s depending on the mass of the star, though their models did not include pulsation–convection coupling. They obtain a trapping-cycle length of ∆k = 5 → 7, and the quadrupole modes showed instabilities comparable to the dipole modes. Buchler, Goupil, & Hansen (1997) show that if there is a resonance between pulsation modes, even if the mode is stable, its amplitude will be necessarily nonzero. They also point out that in case of amplitude saturation, it is the smaller adjacent modes that show the largest amplitude variation, not the main modes. However, if the combination peaks are not real modes in a physical sense, just non-linear distortion by the medium, it is not clear that one would have resonant (mode-coupling) between the combination peaks and real modes. When the Whole Earth Telescope observed GD 358 in 1990, 181 periodicities were detected, but only modes from radial order k = 8 to 18 were identified, most of them showing triplets, consistent with the degree ` = 1 identification. In fact, the observed period spacing is consistent with the measured parallax only if the observed pulsations have degree ` = 1 (Bradley & Winget 1994). Vuille et al. (2000) studied the 342 hours of Whole Earth Telescope data obtained in 1994, showing again modes with k = 8 to 18, and discovered up to 4th-order cross-frequencies in the power spectra. They compared the amplitudes and phases observed with those predicted by the pulsation–convection interaction proposed by Brickhill (1992), and found reasonable agreement. Note that the number of nodes in the radial direction k cannot be determined observationally and rely on a detailed comparison of the observed periods with those predicted by pulsation models.

2. Observations We report here two data sets: 48 hr of time series photometry acquired in August 1996, with the journal of observations presented in Table 1. The second data set consists of 323 hours acquired in May-June 2000 (see Tables 2 and 3 for the observing log). Both of these data sets were obtained simultaneously with time resolved STIS spectroscopy with the Hubble Space Telescope, which will be reported elsewhere. In 1996, the observations were obtained with three channel time series photometry using bi-alkali photocathodes (Kleinman, Nather, & Phillips, 1996), in Texas, China, and Poland. During 23 May to 23 June, 2000, we observed GD 358 mainly with two and three channel time series photometers using bi-alkali photocathodes and a time resolution of 5 s. The May-June 2002 run used 13 telescopes composing the Whole Earth Telescope. The telescopes, ranging from 60 cm to 256 cm in diameter, were located in Texas, Arizona, Hawaii, New Zealand, China, Lithuania, Poland, South Africa, France, Spain, Canary Islands, and Brazil. As the pulsations in white dwarf stars are in phase at different wavelengths (Robinson, Kepler, & Nather 1982), we used no filters, to maximize the detected signal. Each run was reduced and analyzed as described by Nather et al. (1990) and Kepler (1993), correcting for ex- tinction through an estimated local coefficient, and sky variations measured continuously on three channel and CCD observations, or sampled frequently on two channel photometers. The second channel of the photometer monitored a nearby star to assure photometric conditions or correct for small non-photometric conditions. The CCD measurements were obtained with different cameras which are not described in detail here. At least two comparison stars were in each frame and allowed for differential weighted aperture photometry. The consecutive data points were 10 to 30 s apart, depending if the CCDs were frame transfer or not. After this preliminary reduction, we brought the data to the same fractional amplitude scale and converted the middle of integration times to Barycentric Coordinated Time TCB (Standish 1998). We then computed a Discrete Fourier Transform (DFT) for the combined 2000 data, shown in Fig. 1. Due to poor weather conditions during the run, our coverage is not continuous, causing gaps in the observed light curve; these gaps produce aliases in the Fourier transform. At the bottom of Fig. (1) we present the spectral window, the Fourier transform of a single sinusoid sampled exactly as the real data. It shows the pattern of peaks each individual frequency in the data introduces in the DFT. 4 Kepler et al.: The Everchanging Pulsating White Dwarf GD358

Telescope Run Date (UT) Start Time(UT) Length (s) Suhora 60 cm suh-55 Aug 10 23:28:00 8890 Suhora 60 cm suh-56 Aug 12 20:26:40 14730 McDonald 210 cm an-0036 Aug 13 3:06:30 13840 Suhora 60 cm suh-57 Aug 13 19:12:10 22050 McDonald 210 cm an-0038 Aug 14 3:14:10 12920 BAO 85 cm bao-0026 Aug 14 13:10:00 13610 Suhora 60 cm suh-58 Aug 14 23:19:10 5370 McDonald 210 cm an-0040 Aug 15 3:04:30 15780 McDonald 96 cm an-0041 Aug 16 2:54:00 15700 BAO 85 cm bao-0027 Aug 16 13:01:50 1250 BAO 85 cm bao-0028 Aug 16 13:42:30 9330 McDonald 96 cm an-0042 Aug 17 5:01:50 6420 McDonald 96 cm an-0043 Aug 18 4:05:30 3930 Suhora 60-cm suh-59 Aug 18 21:02:10 13010 McDonald 96 cm an-0042 Aug 18 4:05:30 3930 McDonald 96 cm an-0043 Aug 19 2:44:40 2100 McDonald 96 cm an-0044 Aug 19 2:44:40 2100 McDonald 96 cm an-0044 Aug 19 3:47:00 4250 McDonald 96 cm an-0045 Aug 19 3:47:00 4250 McDonald 96 cm an-0046 Aug 19 4:58:30 7160 Suhora 60 cm suh-60 Aug 19 20:28:00 15290 Suhora 60 cm suh-61 Aug 20 21:04:00 10670

Table 1. Journal of ground-based observation for GD358 in August, 1996.

The Fourier spectra displayed in Fig. (1) looks similar to the ones obtained in 1990 and 1994 (see Figure 2), but the amplitude of all the modes changed significantly. As we describe in more detail later, the most striking feature of the 2000 data is the absence of triplets, except for k = 9. The 1996 data are even more unusual than the WET runs due to the observation of amplitude changes over an unprecedented short time. We describe these observations in more detail in section 3. We then describe the 2000 observations and our interpretations of them in Sections 4 through 7.

3. Changes in the Dominant Mode in 1996 We observed GD 358 in August 1996 to provide simultaneous observations to compare to HST time resolved spec- troscopy made on August 16. The 1996 data covers 10 days of the most remarkable amplitude behavior ever seen in a pulsating white dwarf. Observing this behavior is serendipitous, as individual observers and the WET have observed GD 358 off and on for 20 years without seeing this sort of behavior. Figure 3 shows how the amplitude of the k = 8 P = 423 s mode changed with time during our observations in August 1996. The amplitude changes we saw in our optical data are unprecedented in the observations of pulsating white dwarf stars; no report has been made of such a large amplitude variation in such a short amount of time. Here we describe what we found in our data. The lightcurves acquired in August 1996 are displayed in Fig. 4 and the Fourier transform for each lightcurve is shown in Figure 4. Those in the first and second panels of Fig. 4 look very different from each other. The Fourier transform of the lightcurve from the first panel is similar to that from the 1994 WET data where we identified over 100 individual periodicities, while the Fourier transform of the second panel is dominated by only a single periodicity (Fig. 5); this represents a complete change in the mode structure, as well as the period of the dominant mode, in about one day! In run an–0034, the k = 8 P=423 s mode’s amplitude is 170 mma, which is the largest amplitude we have ever seen for this mode. To check for additional pulsation power (perhaps lower amplitude pulsations dwarfed by the 423 s mode power), we prewhitened the an–0034 lightcurve by the 423 s mode. Prewhitening subtracts a sinusoid with a specified amplitude, phase and period from the original lightcurve, and it helps us look for smaller amplitude pulsations by eliminating the alias pattern of the dominant pulsation mode from the Fourier transform. In Fig. 6, we show the Fourier transform of the an–0034 lightcurve both before and after prewhitening. We see now that GD 358 was indeed dominated entirely by a single mode at a different period from the dominant mode a day earlier. We refer to this event by the musical term “forte”, or more informally as the “whoopsie”. Given the spectacular behavior of GD 358 in August 1996, we obtained follow-up observations in September 1996 and April 1997. Table 4 shows the journal of observations for the September 1996 and April 1997 data. The lightcurve Kepler et al.: The Everchanging Pulsating White Dwarf GD358 5

Table 2. Journal of ground-based observation for GD358 in May-June, 2000.

Run Name Telescope Date (UT) Start Time Length (UT) (s) jr0523 Moletai 1.65m May 23 00 22:15:49 6360 tsm-0074 McDonald 2.1m May 24 4:59:00 20050 jr0524 Moletai 1.65m May 24 20:24:35 10255 tsm-0075 McDonald 2.1m May 25 3:25:00 25440 suh-089 SUHORA 0.6m May 25 23:58:30 4715 tsm-0076 McDonald 2.1m May 26 3:18:30 25350 suh-090 Suhora 0.6m May 26 20:07:20 18995 jr0526 Moletai 1.65m May 26 20:26:00 12545 teide02 Tenerife IAC 0.8m May 27 01:25:20 13270 tsm-0077 McDonald 2.1m May 27 03:00:00 26100 suh-091 Suhora 0.6m May 27 19:56:20 15240 sa-et1 SAAO 0.75m May 28 00:18:30 15220 teide04 Tenerife IAC 0.8m May 28 0:37:10 14870 tsm-0078 McDonald 2.1m May 28 2:51:00 29040 teide05 Tenerife IAC 0.8m May 28 21:53:20 24700 jr0528 Moletai 1.65m May 28 20:28:20 9550 tsm-0079 McDonald 2.1m May 29 2:50:00 27300 calto2905 Calar Alto 1.23m May 29 21:20:00 6000 CCD teide06 Tenerife IAC 0.8m May 29 22:01:50 4375 gv-2905 OHP May 29 22:06:00 15530 teide07 Tenerife IAC 0.8m May 29 23:28:00 11460 teide08 Tenerife IAC 0.8m May 30 2:50:50 6850 tsm-0080 McDonald 2.1m May 30 3:00:00 28200 suh-092 Suhora 0.6m May 30 20:39:30 14880 gv-2906 OHP 1.93m May 30 20:44:00 4340 teide09 Tenerife IAC 0.8m May 30 21:58:50 25270 sjk-0401 Hawaii UH 0.6m May 31 7:19:00 26615 gv-2907 OHP, 1.93 m May 31 20:44:00 20400 teide10 Tenerife IAC 0.8m May 31 22:25:20 8195 calto3105 Calar Alto 1.23m May 31 21:30:00 5560 CCD sa-od033 SAAO 0.75m May 31 22:31:00 4000 sjk-0402 Hawaii UH 0.6m Jun 1 6:05:00 31265 jxj-0103 BAO 0.85m Jun 1 13:18:20 6950 suh-093 Suhora 0.6m Jun 1 19:55:30 19310 jr0601 Moletai 1.65m Jun 1 20:12:00 13145 sa-od035 SAAO 0.75m Jun 1 20:22:00 10600 calto0106 Calar Alto 1.23m Jun 1 20:46:00 4845 CCD gv-2908 OHP 1.93m Jun 1 20:48:00 19650 teide11 Tenerife IAC 0.8m Jun 1 22:38:40 2230 teide12 Tenerife IAC 0.8m Jun 1 23:16:20 12460 tsm-0081 McDonald 2.1m Jun 2 2:50:00 24000 teide13 Tenerife IAC 0.8m Jun 2 3:00:20 7180 sara030 Sara 0.9m Jun 2 8:30:00 14400 CCD sjk-0403 Hawaii UH 0.6m Jun 2 6:04:00 32505 jxj-0003 BAAO 0.85m Jun 2 15:24:50 11030 suh-094 Suhora 0.6m Jun 2 19:53:00 19530 gv-2909 OHP 1.93m Jun 2 21:40:00 16750 jr0602 Moletai 1.65m Jun 2 22:03:45 4740 calto0602 Calar Alto 1.23m Jun 2 23:20:00 4030 CCD teide15 Tenerife IAC 0.8m Jun 3 00:22:40 16630

and the power spectrum during these observations (Fig. 7) seem to have returned to the more or less normal state seen in the past (Fig. 2), not the unusually high amplitude state it was in in August 1996 (Fig. 5). We show the lightcurves of GD 358 at three different times in Fig. 8. The middle panel shows the lightcurve when the amplitude of the 423 s mode was at its largest. The lightcurve looks almost sinusoidal, with the single 423 s mode in the power spectrum. The result of this is that we obtain similar values for the peak-to-peak semi-amplitudes and the FT amplitude of the 423 s mode; this implies that a single spherical harmonic is a good representation of the stellar 6 Kepler et al.: The Everchanging Pulsating White Dwarf GD358

Table 3. Journal of ground-based observation for GD358 in May-June, 2000. (cont.)

Run Name Telescope Date (UT) Start Time Length (UT) (s) calto0602.2 Calar Alto 1.23m Jun 3 00:35:00 2800 tsm-0082 McDonald 2.1m Jun 3 3:03:30 22110 sara031 Sara 0.9m Jun 3 4:15:00 26100:30 CCD sjk-0404 Hawaii UH 0.6m Jun 3 5:59:30 31270 jxj-0105 BAO 0.85m Jun 3 13:27:50 4185 sjk-0405 Hawaii UH 0.6m Jun 3 14:42:30 1465 jxj-0106 BAO 0.85m Jun 3 16:00:20 8725 suh-095 Suhora 0.6m Jun 3 20:02:50 19055 jr0603 Moletai 1.65m Jun 3 20:24:55 12305 sa-od037 SAAO 0.75m Jun 3 20:49:00 10655 teide17 Tenerife IAC 0.8m Jun 4 00:26:00 16760 sara032 Sara 0.9m Jun 4 4:08:00 26940 CCD sjk-0406 Hawaii UH 0.6m Jun 4 5:37:00 34055 tsm-0083 McDonald 2.1m Jun 4 7:46:30 11310 jxj-0107 BAO 0.85m Jun 4 12:32:50 25280 suh-096 Suhora 0.6m Jun 4 20:21:00 16775 sa-od039 SAAO 0.75m Jun 4 21:42:00 7110 calto0604 Calar Alto 1.23m Jun 4 23:06:37 4290 CCD teide19 Tenerife IAC 0.8m Jun 5 0:14:30 17270 tsm-0084 McDonald 2.1m Jun 5 3:00:00 6150 sara034 Sara 0.9m Jun 5 4:47:00 24720 CCD jxj-0108 BAO 0.85m Jun 5 12:33:20 25295 suh-097 Suhora 0.6m Jun 5 20:04:00 18545 jr0605 1 Moletai 1.65m Jun 5 20:58:25 5555 sa-od042 SAAO 0.75m Jun 5 21:49:00 8005 jr0605 2 Moletai 1.65m Jun 5 22:52:55 3925 teide20 Tenerife IAC 0.8m Jun 6 1:08:00 13685 tsm-0085 McDonald 2.1m Jun 6 2:55:00 28800 sara035 Sara 0.9m Jun 6 4:08:00 10080 CCD edjoh01 NOT Jun 6 21:33:40 10150 edjoh02 NOT Jun 7 1:33:10 13025 teide22 Tenerife IAC 0.8m Jun 7 0:16:40 16850 teide23 Tenerife IAC 0.8m Jun 7 21:00:00 28800 edjoh03 NOT 2.5m Jun 7 22:23:40 15755 suh-098 Suhora 0.6m Jun 8 20:05:10 17065 sara036 Sara 0.9m Jun 10 4:41:00 7800 CCD sara037 Sara 0.9m Jun 11 4:01:00 9780 CCD sara038 Sara 0.9m Jun 12 3:57:00 10600 CCD sara039 Sara 0.9m Jun 20 7:09:00 14700 CCD sara040 Sara 0.9m Jun 21 3:36:00 28920 CCD sara041 Sara 0.9m Jun 22 3:29:00 22020 CCD sara042 Sara 0.9m Jun 23 3:23:00 28680 CCD

pulsation at this time. The other two lightcurves, however, each containing several pulsation modes, are less sinusoidal. If the non-sinusoidal nature of a lightcurve comes from the fact that many modes are present simultaneously, then one would expect the shape of the lightcurve to be sinusoidal only when it is pulsating in a single mode. On the other hand, in the August 1996 sinusoidal lightcurve, the peak–to–peak light variation was about 44% of the star’s average light in the optical. We would expect such a large light variation to introduce nonlinear effects into the lightcurve, even if the star is pulsating in a single mode, causing the lightcurve to look nonsinusoidal. Thus, the nearly sinusoidal shape of our lightcurves (Fig. 8) is a mystery, except for the theoretical models of Ising & Koester 2001, which predict sinusoidal shapes for large amplitude modes even with the nonlinear response of the envelope. After the P = 423 s mode reached its highest amplitude in run an-0034, the k = 9 P = 464 s mode started to grow and the 423 s became smaller, but there was still very little sign of the usually dominant k = 17 P = 770 s mode. In Fig. 2, we present the Fourier amplitude spectra of the light curves obtained each year, 2000 on top, 1996, 1994, and 1900 on the bottom, on the same vertical scale. Note that the 1996 data set is low resolution, because of its smaller amount of data. It is clear that the periodicities change amplitude from one data set to the other. It is important to Kepler et al.: The Everchanging Pulsating White Dwarf GD358 7

Telescope Run Date (UT) Time(UT) Length (s) PdM 2m gv-0480 1996 Sep 10 20:29:01 5670 Suhora 60cm suh-62 1996 Sep 11 18:11:00 10790 PdM 2m gv-0484 1996 Sep 14 21:22:02 2330 Suhora 60cm suh-63 1996 Sep 18 18:45:00 15860 Suhora 60cm suh-65 1996 Sep 19 18:06:20 13380 McD 2.1m an-0061 1997 Apr 1 06:54:20 415 McD 2.1m an-0066 1997 Apr 7 06:52:50 1763

Table 4. Journal of Observation for September, 1996 and April, 1997. September, 1996 data were taken by our Whole Earth Telescope collaborators during a WET run whose primary target was not GD 358. PdM stands for Pic du Midi in France, and Suhora is for Mt. Suhora in Poland. The 1997 data were all taken at McDonald Observatory in Texas.

notice that the periodicities, when present, have similar frequencies over the years. The amplitudes change, and even subcomponents (different m values) may appear and disappear, but when they are present, they have basically the same frequencies (typically to within 1 µHz). In the September 1996 data, the Fourier transform shows that GD 358 is pulsating in periods similar to what we are familiar with from the WET data of 1990, although the highest peaks are at 1082 µHz, 2175 µHz and 2391 µHz. The very limited data set and the complex pulsating structure of the star makes interpretation of these peaks difficult (Fig. 7). It is not until the data taken in April 1997 when we observe the 770 s mode as the highest amplitude mode in the Fourier transform, as in 1990 and 1994. We do not have data to fill in the gap between September 1996 and April 1997 to see how the amplitude changed, but even by August 19th, the modes at k = 15 and 18 were already starting to appear. The time scale which the star took to change from its normal multi–mode state to a single mode pulsator was very short, about one day. The reverse transition started one week after the event. An estimate of the total energy observed in pulsations is best obtained by measuring the peak-to-peak amplitudes in the light curves directly, instead of adding the total power from all the modes. For the largest amplitude run in 1996, an-0034, observed with the 82” telescope at McDonald, we estimate a peak-to-peak semi-amplitude of 220 mma. For comparison, the measured Fourier amplitude for the k = 8 mode for that run is 170 mma. For two runs at the same telescope in 2000, we obtain a peak-to-peak semi-amplitude of 120 mma. Again for comparison, the Fourier amplitudes of the large modes present are 30 mma, but there are several modes, and many combination peaks. As the observed pulsation energy is quadratic in the amplitudes and the frequencies, it corresponds to an increase of around 34% in the radiated energy by pulsations, from the amplitudes, plus a factor of 2.8 from the frequency. Just two days after the “forte”, the peak-to-peak amplitude decreased by a factor of 5, but during our observations a month later, it had already increased to its pre-“forte” value. It is important to notice that the observed amplitude is not directly a measurement of the physical amplitude, as there are several factors that typically depend on `, including: geometrical cancellation, inclination effects, kinetic energies associated with the oscillatory mass motions, together with a term that depends on the frequency of pulsation squared. If we assume that the inclination angle of the pulsation axis to our line of sight does not change, and that the ` values of the dominant modes do not change, then it must be the ` distribution of the combination frequencies that changes and produces a difference in the peak-to-peak variations in the light curve, if the total energy is conserved. This is plausible, as relatively small variations in the amplitudes of the dominant periods can dramatically change the amplitudes of the linear combination frequencies, but not necessary.

4. Main Periodicities in the 2000 WET Data 4.1. Assumptions and Ground Rules Used in this Work We observed GD 358 as the primary target in May-June of 2000 to provide another “snapshot” of the behavior of GD 358 with minimal alias problems. For the period May 23rd to June 8th, this run provided coverage (∼ 80%) that was intermediate between the 1994 run (86% coverage) and the 1990 run (with 69% coverage). The 2000 WET run had several objectives: 1) look for additional modes besides the known ` = 1 k = 8 through 18 modes; 2) investigate the multiplet splitting structure of the pulsation modes; 3) look for amplitude changes of the known modes; 4) determine the structure of the “combination peaks”, including the maximum order seen; and 5) provide simultaneous observations for HST time resolved spectroscopy. Before we can start interpreting the peaks in the FT, we need to select an amplitude limit for what constitutes a “real” peak versus a “noise” peak. Kepler (1993) and Schwarzenberg-Czerny (1991, 1999), following Scargle (1982), demonstrated that non-equally spaced data sets and multiperiodic light curves, as all the Whole Earth Telescope data sets are, do not have a normal noise distribution, because the residuals are correlated. The probability that a peak in 8 Kepler et al.: The Everchanging Pulsating White Dwarf GD358

Table 5. Average amplitude of datasets, from 1000 to 3000 µHz.

Year BCTstart hAmpi (days) (mma) 1990 244 8031.771867 0.62 1994 244 9475.001705 0.58 1996 245 0307.617884 1.44 2000 245 1702.402508 0.29

the Fourier transform has a 1/1000 chance of being due to noise, not a real signal, for our large frequency range of interest,1 requires at least peaks above 4hAmpi, where the average amplitude hAmpi is the square root of the power average (see also Breger et al. 1993 and Kuschnig et al. 1997 for a similar estimative). Table 5 shows that the noise, represented by hAmpi, for the 2000 data set is the smallest to date, allowing us to detect smaller amplitude peaks. Several peaks in the multi-frequency fits are below the 4hAmpi limit and therefore should be considered only as upper limits to the components. The present mode identification follows that of the 1990 data set, published by Winget et al. (1994). They represent the pulsations in terms of spherical harmonics Y`,m, with each eigenmode described by three quantum numbers: the radial overtone number k, the degree `, also called the angular momentum quantum number, and the azimuthal number m, with 2` + 1 possible values, from −` to +`. For a perfectly spherical star, all (2` + 1) eigenmodes with the same values of k and ` should have the same frequency, but rotation causes each eigenmode to have a frequency also dependent on m. Magnetic fields also lift the m degeneracy. The assigned radial order k value are the outcome of a comparison with model calculations presented in Bradley & Winget (1994), and are consistent with the observed mass and parallax, as discussed in their paper. Vuille et al. (2000) determinations followed the above ones. In the upper part of Fig.(1), we placed a mark for equally spaced periods (correct in the asymptotic limit), using the 38.9 s spacing derived by Vuille et al., starting with the k = 17 mode. The observed period spacings in the FT are very close to equal, consistent with previous observations.

4.2. Nonlinear Least Squares Results from 1990, 1994, and 1996 For a more self-consistent comparison, we took the data from the 1990, 1994 WET runs and the August 1996 run and derived the periods of the dominant modes via a nonlinear least squares fit. In Tables 6, 7 and 8 we present the results of a non-linear simultaneous least squares fit of 23 to 29 sinusoids, representing the main periodicities, to the 1990, 1994 and 1996 data sets. We use the nomenclature ka, for example 15−, to represent a subcomponent with m = −1 of the k = 15 mode in these tables. The difference in the frequencies reported in this paper compared to the previous ones is due to our use of the simultaneous non-linear least-squares frequency fitting rather than using the Fourier Transform frequencies. We note that both the Fourier analysis and multi-sinusoidal fit assume the signal is composed of sinusoids with constant amplitudes, which is clearly violated in the 1996 data set. The changing amplitudes introduce spurious peaks in the Fourier transform. This will not affect the frequency of the modes, but the inferred amplitude will be a poor match to the (non-sinusoidal) light curve amplitude. In Table 5 we present the average amplitude of the data sets, from 1000 to 3000 µHz, after the main periodicities, all above 4hAmpi, have been subtracted. For the 2000 data set, the initial hAmpi for the frequency range from 0 to 10 000 µHz, is 0.69 mma. For the high frequency range above 3000 µHz, hAmpi ' 0.2 mma.

4.3. Mode Analysis of the 2000 WET Data To provide the most accurate frequencies possible, we rely on a non-linear least squares fit of sinusoidal modes with guesses to the observed periods, since these better take into account contamination or slight frequency shifts due to aliasing. In Table 9 we present the results of a simultaneous non-linear least squares fit of 29 sinusoids, representing the main periodicities of the 2000 data set, simultaneously. All the phases have been measured with respect to the barycentric Julian coordinated date BCT 2 451 702.402 508. Armed with the new frequencies in Table 9, we comment on regions of particular interest in the FT. First, we identify several newly detected modes at P = 373.76 s, f = 2675.49 µHz, amp=8.43 mma; P = 852.52 s, f = 1172.99 µHz, amp=2.74 mma; and P = 900.13 s, f = 1110.95 µHz, amp=2.03 mma. Based on the mode assignments of Bradley & Winget (1994) we identify these modes as k = 7, 19, and 20. The mode identification is based on the proximity of

1 corresponding to a number of multiple trials larger than the number of data points Kepler et al.: The Everchanging Pulsating White Dwarf GD358 9

Table 6. Our multisinusoidal fit to the main periodicities in 1990.

k Frequency Amplitude Tmax (µHz) (mma) (s) 18 1233.408 ± 0.017 5.05 ± 0.13 731 ± 7 17+ 1291.282 ± 0.018 5.04 ± 0.14 395 ± 6 17 1297.590 ± 0.006 14.60 ± 0.14 24 ± 2 17− 1303.994 ± 0.019 4.71 ± 0.14 411 ± 7 16+ 1355.664 ± 0.106 0.87 ± 0.14 471 ± 37 16 1361.709 ± 0.040 2.21 ± 0.14 18 ± 14 16− 1368.568 ± 0.031 2.96 ± 0.14 627 ± 11 15+ 1420.932 ± 0.010 9.32 ± 0.14 416 ± 3 15 1427.402 ± 0.005 19.24 ± 0.14 425 ± 2 15− 1433.853 ± 0.011 7.90 ± 0.14 88 ± 4 14+ 1513.023 ± 0.017 5.23 ± 0.14 123 ± 5 14 1518.991 ± 0.009 9.71 ± 0.14 270 ± 3 14− 1525.873 ± 0.016 5.35 ± 0.13 459 ± 5 13+ 1611.671 ± 0.016 5.70 ± 0.14 140 ± 5 13 1617.297 ± 0.017 5.28 ± 0.14 508 ± 5 13− 1623.644 ± 0.019 4.70 ± 0.14 407 ± 5 12 1733.850 ± 0.163 0.53 ± 0.13 474 ± 44 11+ 1840.022 ± 0.136 0.65 ± 0.14 41 ± 35 11 1846.247 ± 0.135 0.66 ± 0.14 504 ± 34 11− 1852.099 ± 0.093 0.94 ± 0.14 132 ± 24 10+ 1994.240 ± 1.071 ≤ 0.14 10 1998.919 ± 0.060 1.50 ± 0.14 25 ± 14 10− 2007.992 ± 0.117 0.84 ± 0.14 7 ± 26 9+ 2150.430 ± 0.048 1.932 ± 0.13 174 ± 10 9 2154.052 ± 0.020 4.59 ± 0.13 336 ± 4 9− 2157.834 ± 0.032 2.81 ± 0.13 400 ± 7 8+ 2358.975 ± 0.016 5.68 ± 0.13 120 ± 3 8 2362.588 ± 0.016 5.77 ± 0.13 422 ± 3 8− 2366.418 ± 0.017 5.34 ± 0.13 268 ± 3

the detected modes with those predicted by the models, or even the asymptotical period spacings, but also because of resonant mode coupling, i.e., a stable mode will be driven to visibility if a coupled mode falls near its frequency, as it happens for k = 7, which is very close to the combination of k = 17 and k = 16, and k = 20, which falls near the resonance of the 8− and the ` = 2 mode at 1255 µHz (see next paragraph). It is important to note that these modes appear in combination peaks with other modes, as shown in Table 11. This reinforces our belief that these modes are physical modes, and not just erroneously identified combination peaks. We note that Bradley (2002) analyzed single site data taken over several years, and found peaks at 1172 or 1183 µHz in April 1985, May 1986, and June 1992 data, lending additional credence to the detection of the k = 19 mode or its alias. The first previously known region of interest surrounds the k = 18 mode, which lies near 1233 µHz, according to previous observations. In the 2000 data, the largest amplitude peak in this region lies at 1255 µHz, which is over 20 µHz from the previous location. Given that other modes (especially the one at k = 17) has shifted by less than 4 µHz, we are inclined to rule out the possibility that the k = 18 mode shifted by 20 µHz. One possible solution is offered by seismological models of GD 358, which predict an ` = 2 mode near 1255 µHz. For example, the best ML2 fit to the 1990 data (from Metcalfe, Salaris & Winget 2002, Table 3), has an ` = 2 mode at 1252.6 µHz (P=798.3 s). This would also be consistent with the larger number of subcomponents detected, although they may be caused only by amplitude changes during the observations. Fig (9) shows the region of the k = 18 mode in the FT for the 1990 data set (solid) and the 2000 data set (dashed); it is consistent with the k = 18 mode being the 1233 µHz for both data sets, and they even have similar amplitudes. While we avoided having to provide an explanation for why only the k = 18 mode would shift by 20 µHz, we have introduced another problem, as geometrical cancellation for an ` = 2 mode introduces a factor of 0.26 in relation to unity for an ` = 1 mode. Thus, the identification of the 1255 µHz as an ` = 2 mode, which has a measured amplitude of 14.86 mma, implies a physical amplitude higher than that of the highest amplitude ` = 1 mode, around 30 mma. Kotak et al. (2002), analyzing time-resolved spectra obtained at the Keck in 1998, show the velocity variations of the k = 18 mode at 1233 µHz is similar to those for the k = 15 and k = 17 modes, concluding all modes are ` = 1. They did not detect a mode at 1255 µHz. 10 Kepler et al.: The Everchanging Pulsating White Dwarf GD358

Table 7. Main periodicities in 1994.

k Frequency Amplitude Tmax (µHz) (mma) (s) 18+ 1228.712 ± 0.022 2.77 ± 0.13 252.6 ± 12.0 18 1235.493 ± 0.005 12.94 ± 0.13 170.9 ± 2.6 18− 1242.364 ± 0.016 3.66 ± 0.13 62.4 ± 9.0 17+ 1291.093 ± 0.010 6.17 ± 0.13 250.3 ± 5.1 17 1297.741 ± 0.003 22.11 ± 0.13 37.7 ± 1.4 17− 1304.459 ± 0.010 6.25 ± 0.13 615.0 ± 5.0 16+ 1355.388 ± 0.035 1.70 ± 0.13 167.0 ± 17.6 16 1362.298 ± 0.060 < 0.89 16− 1368.541 ± 0.031 1.92 ± 0.13 322.3 ± 15.5 15+ 1419.650 ± 0.003 18.37 ± 0.13 46.4 ± 1.6 15 1426.408 ± 0.004 15.55 ± 0.13 239.7 ± 1.8 15a 1430.879 ± 0.006 10.61 ± 0.13 187.9 ± 2.7 15− 1433.169 ± 0.014 4.46 ± 0.13 104.1 ± 6.5 14 1519.903 ± 0.028 1.09 ± 0.13 485.2 ± 20.0 13+ 1611.357 ± 0.012 5.02 ± 0.13 466.0 ± 5.0 13 1617.474 ± 0.009 3.46 ± 0.13 183.3 ± 1.1 13− 1624.568 ± 0.010 6.07 ± 0.13 101.0 ± 4.1 12 1746.766 ± 0.064 0.93 ± 0.13 414.2 ± 25.0 11 1863.004 ± 0.184 < 0.71 10 2027.325 ± 0.457 < 0.46 9+ 2150.504 ± 0.019 3.15 ± 0.13 346.7 ± 6.1 9 2154.124 ± 0.013 4.76 ± 0.13 12.3 ± 4.0 9− 2157.841 ± 0.022 2.69 ± 0.13 144.2 ± 7.1 8+ 2358.883 ± 0.013 4.50 ± 0.13 398.2 ± 3.8 8 2362.636 ± 0.006 9.25 ± 0.13 274.6 ± 1.8 8− 2366.508 ± 0.007 4.22 ± 0.13 169.7 ± 3.8

In Fig. 10 we show the pre-whitened results; pre-whitening was done by subtracting from the observed light curve a synthetic light curve constructed with a single sinusoid with frequency, amplitude and phase that minimizes the Fourier spectrum at the frequency of the highest peak. A new Fourier spectrum is calculated and the next dominant frequency is subtracted, repeating the procedure until no significant power is left. It is important to notice that with pre-whitening, the order of subtraction matters. As an example, in the 2000 data set, if we subtract the largest peak in the region of the k = 18 mode, at 1255.41 µHz, followed by the next highest peak at 1256.26 µHz and the next at 1254.44 µHz, we are left with a peak at only 1.3 mma at 1232.76 µHz. But if instead we subtract only the 1255.41 µHz followed by the peak left at 1233.24 µHz, its amplitude is around 3.1 mma, i.e., larger. Pre-whitening assumes the frequencies are independent in the observed, finite, data set. If they were, the order of subtraction would not affect the result. Because the order of subtraction matters, the basic assumption of pre-whitening does not apply. We attempt to minimize this effect by noting that the frequencies change less than the amplitudes, and use the FT frequencies in a simultaneous non-linear least squares fit of all the eigenmode frequencies. But even the simultaneous non-linear least squares fit uses the values of the Fourier transform as starting points, and could converge to a local minimum of the variance instead of the global minimum. The modes with periods between 770 and 518 s (k = 17 through 13) are present in the 2000 data, though with different amplitudes than in previous years. Another striking feature of the peaks in 2000 is that one multiplet member of each mode has by far the largest amplitude, so that without data from previous WET runs, we would not know that the modes are rotationally split. The frequencies of these modes are stable to about 1 µHz or less with the exception of the 16− mode, where the frequency jumped from about 1368.5 µHz in 1990 and 1994 to about 1379 µHz in 1996 and 2000 (see Fig. 11). Most of these frequency changes are larger than the formal frequency uncertainty from a given run (typically less than 0.05 µHz), so there is some process in GD 358 that causes the mode frequencies to “wobble” from one run to the next. We speculate that this may be related to non-linear mode coupling effects. Whatever the origin of the frequency shifts, it renders these modes useless for studying evolutionary timescales through rates of period change. The k = 12 through 10 modes deserve separate mention because their amplitudes are always small; between 1990 and 2000, the largest amplitude peak was only 1.6 mma. The small amplitudes can make accurate frequency determinations difficult, and all three modes have frequency shifts of 13 to 33 µHz between the largest amplitude peaks in a given mode. The k = 10 mode shows the largest change with the 1990 data showing the largest peaks at Kepler et al.: The Everchanging Pulsating White Dwarf GD358 11

Table 8. Main modes in 1996.

k Frequency Amplitude Tmax (µHz) (mma) (s) 19 1172.66 ± 0.15 2.5 ± 0.6 17.4 ± 47.5 18 1253.65 ± 1.01 < 2.1 17+ 1291.13 ± 0.11 4.3 ± 0.6 653.6 ± 28.1 170 1295.38 ± 0.17 2.6 ± 0.6 139.7 ± 43.6 17− 1304.68 ± 0.11 4.8 ± 0.6 346.3 ± 25.7 16+ 1355.21 ± 2.02 < 1.9 160 1362.55 ± 0.15 2.7 ± 0.6 172.2 ± 41.2 16− 1379.64 ± 0.16 2.7 ± 0.6 378.6 ± 41.0 150 1427.47 ± 0.92 < 2.2 15− 1434.36 ± 0.18 2.2 ± 0.6 154.5 ± 47.1 14 1520.58 ± 0.18 2.0 ± 0.6 398.3 ± 46.6 13+ 1611.60 ± 0.18 2.1 ± 0.6 461.8 ± 42.7 130 1617.51 ± 0.35 1.1 ± 0.6 183.5 ± 79.9 13− 1619.63 ± 0.84 < 2.2 12 1736.10 ± 0.34 1.1 ± 0.6 323.6 ± 75.2 11 1862.93 ± 0.39 0.9 ± 0.6 58.8 ± 78.7 10 2027.41 ± 0.26 1.4 ± 0.6 471.9 ± 47.9 9+ 2149.97 ± 0.07 5.6 ± 0.6 69.8 ± 11.8 90 2153.84 ± 0.05 7.6 ± 0.6 155.2 ± 8.5 9− 2157.89 ± 0.04 9.1 ± 0.6 426.7 ± 7.2 8+ 2358.63 ± 0.03 12.6 ± 0.6 125.3 ± 4.7 80 2362.50 ± 0.02 23.2 ± 0.6 104.8 ± 2.5 8− 2365.98 ± 0.02 22.2 ± 0.6 142.9 ± 2.6

Table 9. Main modes in 2000.

k Frequency Period Amplitude Tmax (µHz) (sec) (mma) (s) 20 1110.960 ± 0.017 900.122 ± 0.014 2.04 ± 0.08 870.19 ± 8.61 19 1172.982 ± 0.013 852.528 ± 0.009 2.74 ± 0.08 164.86 ± 6.07 ` = 2 1255.400 ± 0.002 796.556 ± 0.002 14.86 ± 0.08 747.02 ± 1.05 18 1233.595 ± 0.018 810.639 ± 0.012 1.96 ± 0.08 354.91 ± 8.11 17+ 1294.284 ± 0.094 772.628 ± 0.056 0.38 ± 0.082 579.63 ± 39.54 17 1296.599 ± 0.001 771.248 ± 0.001 29.16 ± 0.08 247.81 ± 0.52 17− 1301.653 ± 0.053 768.254 ± 0.031 0.68 ± 0.08 314.52 ± 22.52 16 1362.238 ± 0.159 734.086 ± 0.086 0.42 ± 0.12 263.36 ± 63.93 16− 1378.806 ± 0.007 725.265 ± 0.004 5.35 ± 0.08 514.70 ± 2.66 15+ 1420.095 ± 0.001 704.178 ± 0.001 29.69 ± 0.08 418.14 ± 0.49 15 1428.090 ± 0.052 700.236 ± 0.025 0.70 ± 0.08 217.29 ± 19.78 15− 1432.211 ± 0.036 698.221 ± 0.018 1.08 ± 0.089 355.58 ± 13.30 14 1519.811 ± 0.134 657.977 ± 0.058 0.266 ± 0.08 301.55 ± 48.28 13+ 1611.084 ± 0.116 620.700 ± 0.045 0.31 ± 0.08 448.66 ± 39.81 13 1617.633 ± 0.174 618.187 ± 0.066 0.21 ± 0.08 448.19 ± 58.97 13− 1625.170 ± 0.235 615.320 ± 0.089 0.15 ± 0.08 299.24 ± 79.32 12 1736.277 ± 0.034 575.945 ± 0.011 1.04 ± 0.08 230.37 ± 10.79 11 1862.871 ± 0.042 536.806 ± 0.012 0.84 ± 0.08 503.66 ± 12.43 10 2027.008 ± 0.028 493.338 ± 0.007 1.29 ± 0.08 350.14 ± 7.49 9+ 2150.462 ± 0.012 465.016 ± 0.003 2.96 ± 0.08 35.49 ± 3.09 9 2154.021 ± 0.007 464.248 ± 0.001 5.34 ± 0.08 252.38 ± 1.71 9− 2157.731 ± 0.014 463.450 ± 0.003 2.57 ± 0.08 174.68 ± 3.53 8+ 2359.119 ± 0.006 423.887 ± 0.001 5.57 ± 0.08 166.60 ± 1.49 8 2362.948 ± 0.094 423.200 ± 0.017 0.38 ± 0.08 81.50 ± 21.93 8− 2366.266 ± 0.006 422.607 ± 0.001 5.79 ± 0.08 418.36 ± 1.44 2 × 18 2510.761 ± 0.021 398.286 ± 0.003 1.70 ± 0.08 334.47 ± 4.58 2 × 17 2593.208 ± 0.004 385.623 ± 0.001 7.82 ± 0.08 249.57 ± 0.96 7 2675.487 ± 0.004 373.764 ± 0.001 8.49 ± 0.08 193.30 ± 0.86 2 × 15 2840.195 ± 0.008 352.089 ± 0.001 4.29 ± 0.08 47.50 ± 1.60 12 Kepler et al.: The Everchanging Pulsating White Dwarf GD358

1998.7 µHz and 2008.2 µHz, while the 2000 data has one peak dominating the region at 2027.0 µHz. An examination of the data in Bradley (2002) shows that the k = 12 mode seems to consistently show a peak near 1733 to 1736 µHz, and that only the 1994 data has the peak shifted to 1746.8 µHz, suggesting that 1994 data may have found an alias peak or that the 1736 µHz mode could be the m = +1 member and the 1746.8 µHz mode is the m = −1 member. The data in Bradley (2002) do not show convincing evidence for the k = 11 or 10 modes, so we cannot say anything else about them. It is interesting to note that the k = 8 and k = 9 modes are always seen as a triplet, with 3.58 µHz separation for k = 9, even in the 1996 data set. Our measured spacings are 3.54 and 3.69 µHz, from m = −1 to m = 0 and m = 0 to m = 1. The k = 8 mode in 2000 shows an m=0 component below our statistical detection limit (A=0.41 mma, when the local hAmpi = 0.29 mma), but the m = 1 and m = −1 modes remain separated by 2 × 3.58 µHz. All the higher k modes are seen as singlets in the 2000 data set. We also note that the k = 8 and 9 modes have by far the most stable frequencies. The frequencies are always the same to within 0.3 µHz, and in some cases better than 0.1 µHz. However, the frequency shifts are large enough to mask any possible signs of evolutionary period change, as Fig. 12 shows. Thus, we are forced to conclude that GD 358 is not a stable enough “clock” to discern evolutionary rates of period change.

5. Multiplet Splittings As pointed out by Winget et al. (1994) and Vuille et al. (2000), the observed triplets in the 1900 and 1994 data sets had splittings ranging from 6.5 µHz from the “external” modes (such as k = 17) to 3.6 µHz for the “internal” modes k = 8 and 9. Winget et al. interpreted these splittings to be the result of radial differential rotation, and Kawaler et al. (1999) examined this interpretation in more detail. An examination of the frequencies found in the 2000 data set, shown in Table 9, shows that the multiplet structure is much harder to discern, since the k = 10 through 20 modes typically have only one multiplet member with a large amplitude. The obvious multiplet members have frequencies that agree with the 1990 data, except for the 16− mode, where there is a +10.234 µHz shift in the 2000 data. Note that the k = 10 mode identified at 2027 µHz is different than the 1994 µHz identified by Winget et al. (1994) in the 1990 data. However, the peak they identified is not the highest peak in that region of the Fourier transform (see Fig. 13). Our analysis of the 1990 data has statistically significant k = 10 peaks close to 1999 µHz and 2008 µHz. The only modes with obvious multiplet structure are the k = 9 mode, which still shows an obvious 3.6 µHz split triplet, and the k = 8 mode, which shows two peaks that are consistent with 2 × 3.6 µHz separation. In Table 11 we have a peak 3.3 µHz from the k = 15, m = 1 mode that we have not seen before; we call it the 15b mode. We are not certain whether this is another member of the k = 15 multiplet (analogous to the 15a mode in the 1994 WET data) or if it is something else. The 1994 data set also presented a large peak 4.4 µHz from k = 15, m = 0, which we call the 15a mode, in addition to the m = ±1 components. We have not seen this 15a mode in any other data set other than the 1994 WET run. The identity of the “extra subcomponents” remains an unsolved mystery.

6. Linear Combinations Winget et al. (1994) and Vuille et al. (2000) show that most of the periodicities are in fact linear combination peaks of the main peaks (eigenmodes). Combination peaks are what we call peaks in the FT whose frequencies are equal to the sum or difference of two (or more) the the ` = 1 or 2 mode frequencies. The criteria for selection of the combination peaks was that the frequency difference between the combination peak and the sum of the “parent mode” frequencies must be smaller than our resolution, which is typically around 1 µHz. The last column of Table 11 shows the frequency difference. **** TABLE 11 - Linear combinations (should be here) *** For example, only 28 of the more than 180 peaks in Winget et al. (1994) are ` = 1 modes; the rest are combination peaks up to third order (i.e., three modes are involved). The ` = 1 modes lie in the region 1000 to 2500 µHz, and are identified as modes k = 18 to 8. In the 1994 data set analyzed by Vuille et al., combination peaks up to 4th order were detected. In the 2000 data set we identify combination peaks up to 6th order, and most if not all remaining peaks are in fact linear combination peaks, as demonstrated in Table 11 and is shown in the pre-whitened FT of the 2000 data (see Fig. 14). Here too, we use the nomenclature ka, for example 15−, to represent a subcomponent with m = −1 of the k = 15 mode. The so-called ` = 2 mode at 1255 µHz, as well as k = 17 and k = 15 modes, have subcomponents, but probably they are not different m value components, and are caused, most likely, by amplitude modulation. We say this because the frequency splittings are drastically different than in previous data, and for the ` = 2 mode, there are more than 5 possible subcomponent peaks present. We did not do an exhaustive search for all of the possible combination peaks up to 6th order in the Fourier transform, as we only took into account the peaks that had a probability smaller than 1/1000 of being due to noise, and studied if they could be explained as combination peaks. Kepler et al.: The Everchanging Pulsating White Dwarf GD358 13

Table 10. Optimal Fits to 2000 data.

Parameter Fit a Fit b Fit c Fit d

Teff (K) 24,300 23,500 24,500 22,700 M∗(M¯) 0.61 0.60 0.625 0.630 log(MHe/M∗) -2.79 -5.13 -2.58 -4.07 XO 0.81 0.99 0.39 0.37 q(m/M∗) 0.47 0.47 0.83 0.42 σP (s) 2.60 3.65 2.12 1.72 σ∆P (s) 4.07 4.92 2.21 ···

Brickhill’s (1992) pulsation–convection interaction model predicts, and the observations reported by Winget et al. and Vuille et al. agree, that a combination peak involving two different modes always has a larger relative amplitude than a combination involving twice the frequency of a given mode (also called a harmonic peak). Wu’s (2001) analytical expression leads to a factor of 1/2 difference between a combination peak with two modes versus a harmonic peak, assuming that the amplitudes of both eigenmodes are the same. Vuille et al. claim that the relatively small amplitude of the k = 13 mode in 1994 is affected by destructive beating of the nonlinear peak (2 × 15 − 18) and that the k = 16 mode amplitude is affected by the (15 + 18 − 17) combination peak. It is noteworthy that the peak at 1423.62 µHz is only 3.52 µHz from k = 15, so it might be the 15− mode. However, the previously identified 15− was 6.7 µHz from it, and we consider the 1423.62 µHz peak to be either a result of amplitude modulation of the k = 15 mode or yet another combination peak. We note that the wealth of combination peaks and their relative amplitude offers insight into the amplitude limiting mechanism and would be worthy of the considerable theoretical and numerical effort required to understand it.

7. Model-Fitting with a Genetic Algorithm One of the major goals of our observations of GD 358 was to discover additional modes to help refine our seismological model fits. We were also interested in how much the globally optimal model parameters would change due to the slight shifts in the observed periods. With these goals in mind, we repeated the global model-fitting procedure of Metcalfe, Winget & Charbonneau (2001) on several subsets of the new observations. Our model-fitting method uses the parallel genetic algorithm described by *** Metcalfe & Charbonneau (2002) to minimize the root-mean-square (rms) differences between the observed and calculated periods (Pk) and period spacings (∆P ≡ Pk+1 − Pk) for models with effective temperatures (Teff ) between 20,000 and 30,000 K, total stellar masses (M∗) between 0.45 and 0.95 M¯, helium layer masses with − log(MHe/M∗) between 2.0 and ∼7.0, and an internal C/O profile with a constant oxygen mass fraction (XO) out to some fractional mass (q) where it then decreases linearly in mass to zero oxygen at 0.95 m/M∗. This technique has been shown to find the globally optimal set of parameters consistently among the many possible combinations in the search space, but it requires between ∼200 and 4000 times fewer model evaluations than an exhaustive search of the parameter-space to accomplish this, and has a failure rate < 10−5. We attempted to fit the 13 periods and period spacings defined by the m = 0 components of the 14 modes identified as k = 7 to k = 20 in Table 9. Because of our uncertainty about the proper identification of k = 18 (see section 4) we performed fits under two different assumptions: for Fit a we assumed that the frequency near 1233 µHz was k = 18 (similar to the frequency identified in 1990), and for Fit b we assumed that the larger amplitude frequency near 1255 µHz was k = 18. The results of these two fits led us to prefer the identification for k = 18 in Fit a, and we included this in an additional fit using only the 11 modes from k = 8 to k = 18, which correspond to those identified in 1990 (Fit c). We performed an additional fit (Fit d) that included the same 13 periods used for Fit a, but ignored the period spacings. The optimal values for the five model parameters, and the root-mean-square residuals between the observed and computed periods (σP ) and period spacings (σ∆P ) for the four fits are shown in Table 10. Our preferred solution from Table 10 is Fit a, because it includes our favored identification for the k = 18 mode and the additional pulsation periods. The larger σ∆P in Fit a compared to Fit c is dominated by the large period spacings between the k = 19 and 20 modes (47.6 s) and the k = 7 and 8 modes (49.4 s). Fit a has a mass and effective temperature that are essentially the same as the fit of Bradley & Winget (1994), and are consistent with the spectroscopic values derived by Beauchamp et al. (1999). The other structural parameters are otherwise similar to those found by Metcalfe et al. (2001) (Teff = 22, 600 K, M∗ = 0.650M¯, log(MHe/M∗ = −2.74, XO = 0.84, and q = 0.49). We caution, however, that the large values of σP and σ∆P for Fit a imply that our model may not be an adequate representation of the real white dwarf star. New and unmodeled physical circumstances may have arisen between 1994 and 2000 (e.g. whatever caused the forte in 1996), which may account for the diminished capacity of our simple model to match the observed periods. 14 Kepler et al.: The Everchanging Pulsating White Dwarf GD358

8. Summary of Observational Clues Before embarking on our discussion, we recap the highlights of our observations. First, the 2000 WET data shows eigenmodes from at least k = 8 through k = 19. We may also have detected the k = 7 and k = 20 modes. However, their frequencies are similar to that of unrelated combination peaks, so their identification is less secure. For the first time, we have also found an ` = 2 mode in the GD 358 data; it is at 1255.4 µHz. Second, we see relatively few multiplet modes for a given k, with the exception of the k = 8 and 9 modes. While the multiplet structure of the ` = 1 modes is muted, the combination peaks are enhanced to the point that we see combination modes up to 6th order. Combined with the previous WET runs, we see evidence for anticorrelation between the presence of multiplet structure and combination peaks. The presence of amplitude variability of the ` = 1 mode continues. In the August 1996 data, we saw the most extreme example yet, where all of the observed light was in a single (k = 8 mode) for one night (which we call the “forte”). Data before and after the run show power in the nights before and after in other pulsation modes besides the k = 8 and a much lower amplitude. The periods from the 1996 data are consistent with the 2000 data set, although there are differences in the details.

9. Discussion and Puzzles Using the k = 7, 19, and 20 modes in seismological fits produces a best-fitting model that is similar to that derived from only the k = 8 through 18 modes, indicating that the new modes do not deviate drastically from the expected mode pattern. The reappearance of modes with frequencies similar to those obtained before the mode disappeared (true of all modes from k = 8 through 19), shows that the stellar structure sampled by these modes remained the same for almost 20 years. This is in spite of rapid amplitude change events like the “forte” one observed in August 1996. Our observations, coupled with guidance from the available theories of Brickhill (1992) and Wu & Goldreich (2001) suggest that the “forte” event was probably an extreme manifestation of a nonlinear mode-coupling event that did not materially affect the structure of the star other than possibly the driving region. The appearance and disappearance of modes is similar to the behavior observed in the ZZ Ceti star G 29-38 by Kleinman et al. (1998), and we note that “ensemble” seismology works for GD 358 as well as for the cool ZZ Ceti stars. The one caveat is that the ∼ 1 µHz frequency “wobbles” will place a limit on the accuracy of the seismology. We also appear to have discovered an ` = 2 mode (at 1255.4 µHz) in GD 358 for the first time, based on the match of the observed period to that of ` = 2 modes from our best fitting model. Our model indicates that this is the k = 34 mode. This mode has a relatively large amplitude of 14.9 mma, which combined with the increased geometric cancellation (about 3.8×) of an ` = 2 mode, implies that it has the largest amplitude of any mode observed in 2000. We note the existence of several linear combination peaks involving the 1255 µHz mode, that also show complex structures. This lends credence to the 1255 µHz mode being a real mode, and that the complex structure is associated with the real mode (such as amplitude modulation), as opposed to being some sort of combination peak. The amplitude of the 1255 µHz mode changed during the WET run, so we suspect that the many subcomponents observed are most likely due to amplitude modulation. The period structure of the 1990 and 1994 WET data sets are similar, but show that the amplitude of the modes, and even the fine structure, changes with time. In August 1996, the period structure changed rapidly and dramatically, with essentially all the observed pulsation power going to the k = 8 mode. In spite of the large amplitude, the light curve was surprisingly sinusoidal, with a small contribution from the k = 9 mode. Single site observations one month earlier (June 1996) and one month later (September 1996) show a period structure similar to those present in the 1990 and 1994 data sets. For the 2000 data set, the period structure shows close to equal frequency splittings, and the fine structure is different than observed before. Only the k = 9 mode show the same clear triplet observed in 1990 and 1994, with the same frequency splitting. The k = 8 mode shows the m = −1 and m = 1 modes, while the central m = 0 mode is below our 4hAmpi significance level. The other modes do not show clearly the triplet structure previously observed. The 1990 and 1994 data sets show the m-splitting expected by rotational splitting, but the change of the splitting frequency difference from 6 µHz to 3 µHz from k = 17 to k = 8 was interpreted as indicating differential rotation. The apparent anticorrelation between the abundance of multiplet structure and the highest order of combination frequencies seen is a puzzle. As we do not expect the differential rotation profile of GD 358 changed in the last 10 years (and the splittings we do see in 2000 support this contention), the anticorrelation must be telling something about what is going on with rotation in the convection zone. We say this because the combination peaks are believed to be caused by the nonlinear response of the depth varying convection zone, and thus the increased order of combination peaks implies that the convection zone is more “efficient” at mixing eigenmodes to observable amplitudes. The k = 8 and 9 modes continue to show obvious multiplet structure and little, if any, change in splitting. These modes are the most “internal” of the observed modes of GD 358, and we speculate that this must have some bearing on their Kepler et al.: The Everchanging Pulsating White Dwarf GD358 15 multiplet structure’s ability to persist. We do not see any obvious pattern in the dominant amplitude multiplet member with overtone number, so there is not an obvious pattern of rotational coupling to the convection zone for determining mode amplitude. We will need theoretical guidance to make sense of these observations. A related puzzle is the presence of extra multiplet members and/or apparent large frequency shifts of modes in the k = 15 and 16 multiplets. The k = 15 mode shows an extra component at 1430.88 µHz in the 1994 data and a peak at 1423.62 µHz in the 2000 data that have not been seen before or since. Some possible explanations include: rapid amplitude modulation of a k = 15 multiplet member that the FT interprets as an extra peak; the 2000 peak is about the right frequency to be another ` = 2 mode, if we use the 1255.4 µHz mode as a reference point; it could be an unattributed combination peak involving sums and differences of known modes; or it could be something else entirely. The large peak at about 1379 µHz in 1996 and 2000 is also something of a mystery. It is possible that the k = 16, m = −1 component really changed by 10 µHz from the 1368 µHz observed in 1990, although we would have to explain why only this large amplitude multiplet member suffered this large a frequency change. Other possibilities include: the peak is a 1 cycle per day alias of another mode; the peak is a combination peak — the combination 15 + (` = 2) − 17 is a perfect frequency match; or possibly an ` = 2 mode, based on period spacing arguments. Further observations, data analysis with tools like wavelet analysis, and further model fitting may help determine which explanation fits the data best. Brickhill (1992) proposed that the combination frequencies result from mixing of the eigenmode signals by a depth- varying surface convection zone when undergoing pulsation. He pointed out that the convective turnover time in DA and DB variable white dwarf stars occurs on a timescale much shorter than the pulsation period. As a consequence, the convective region adjusts almost instantaneously during a pulsation cycle. Brickhill demonstrated that the flux leaving the convective zone depends on the depth of the convective zone, which changes during the pulsation cycle, distorting the observed flux. This distortion introduces combination frequencies, even if the pulsation at the bottom of the convection zone is linear, i.e., a single sinusoidal frequency. Wu (2001) analytically calculated the amplitude and phases expected of such combination frequencies, and concluded that the convective induced distortion was roughly in agreement with GD358’s 1994 observations, provided that the inclination of the pulsation axis to the line of sight is between 40 deg and 50 deg. Wu also calculated that the harmonics for ` = 2 modes should be much higher than for ` = 1. However the theory overpredicts the amplitude of the ` = 1 harmonics. She also predicts that geometrical cancellation will, in principle, allow a determination of ` if both frequencies sums and differences are observed. These predictions still need testing. While Wu & Goldreich (2001) discuss parametric instability mechanisms for the amplitude of the pulsation modes, they only discuss the case where the parent mode is unstable and the daughter modes are stable. However, with GD 358, we have a different situation. The highest frequency k = 8 and 9 modes can have as a daughter mode one of the lower frequency (k = 17, 18, or 19) ` = 1 modes and an ` = higher mode. One or both or these daughter modes are actually pulsationally unstable as well, which we believe would require coupling to still lower frequency granddaughter modes that are predicted to be stable by our models and the calculations of Brickhill (1990, 1991) and Goldreich & Wu (1999a, b). We suggest that occasionally the nonlinear coupling of the granddaughter and daughter modes with the k = 8 and 9 modes can allow the k = 8 and 9 modes to suffer abrupt amplitude changes when everything is “just right”. In the meantime, the granddaughter modes will couple to the excited daughter modes (k = 13 through 19 in general) to produce the observed amplitude instability of these modes. We need a quantitative theoretical treatment of this circumstance worked out to see if the predicted behavior matches what we observe in GD 358. Observations of GD 358 have been both rewarding and vexing. We have been rewarded with enough ` = 1 modes being present to decipher the mode structure and perform increasingly refined asteroseismology of this star, starting with Bradley & Winget (1994) up to the latest paper of Metcalfe et al. (2002). One thing asteroseismology has not provided us with is the structure of and/or the depth of the surface convection zone. This would help us test the “convective driving” mechanism introduced by Brickhill (1991) and elaborated on by Goldreich & Wu (1999a,b). Our observations point out the need for further refinements of the parametric instability mechanism described by Wu & Goldreich (2001) to better cover the observed mode behavior. The observational data set is quite rich, and coupled with more detailed theories, offers the promise of being able to unravel the mysteries of amplitude variation seen in the DBV and DAV white dwarfs. This in turn, may offer us the insights needed to ascertain why only some of the predicted modes are seen at any one time. MAW, AKJ, AEC, and MLB acknowledge support by the National Science Foundation through the Research Experiences for Undergraduates Summer Site Program to Florida Tech.

References

Althaus, L. G. & Benvenuto, O. G. 1997, MNRAS, 288, L35 Beauchamp, A., Wesemael, F., Bergeron, P., Fontaine, G., Saffer, R. A., Liebert, J., & Brassard, P. 1999, ApJ, 516, 887 16 Kepler et al.: The Everchanging Pulsating White Dwarf GD358

Bradley, P. A. 2002, in “Radial and Nonradial Pulsations as Probes of Stellar Physics”, eds. C. Aerts, T.R. Bedding, and J. Christensen-Dalsgaard, ASP Conference Series, 259, 382 Bradley, P. A. & Winget, D. E. 1994, ApJ, 430, 850 Brassard, P., & Fontaine, G. 2002, in White Dwarfs, ed. Martino, D., Silvotti, R. , Solheim, J.-E., & Kalytis, R. (Dordretch: Kluwer), in press. Breger, M. et al. 1993, A&A, 271, 482 Brickhill, A. J. 1992, MNRAS, 259, 529 Brickhill, A. J. 1991, MNRAS, 251, 673 Brickhill, A. J. 1990, MNRAS, 246, 510 Buchler, J. R., Goupil, M.-J., & Hansen, C. J. 1997, A&A, 321, 159 Canuto, V. M., Goldman, I., & Mazzitelli, I. 1996, ApJ, 473, 550 Clemens et al. 2002 Clemens, J. C., van Kerkwijk, M. H., & Wu, Y. 2000, MNRAS, 314, 220 Dehner, B. T. & Kawaler, S. D. 1995, ApJ, 445, L141 Fontaine, G. & Brassard, P. 2002, ApJ, 581, L33 Gautschy, A. & Althaus, L. G. 2002, A&A, 382, 141 Goldreich, P. & Wu, Y. 1999a, ApJ, 511, 904 Goldreich, P. & Wu, Y. 1999b, ApJ, 523, 805 Ising, J. & Koester, D. 2001, A&A, 374, 116. Kawaler, S. D., Sekii, T., & Gough, D. 1999, ApJ, 516, 349 Kepler, S. O. 1984, ApJ, 286, 314 Kepler, S. O. 1993, Baltic Astronomy, 2, 515 Kleinman, S. J., Nather, R. E. & Phillips, T. 1996, PASP, 108, 356 Kleinman, S. J. et al. 1998, ApJ, 495, 424 Kotak, R., van Kerkwijk, M. H., Clemens, J. C., & Koester, D. 2002, ˚a,in press (astroph 021691) Kuschnig, R., Weiss, W. W., Gruber, R., Bely, P. Y., & Jenkner, H. 1997, A&A, 328, 544 Metcalfe, T.S. & Charbonneau, P. 2003, J. Comp. Phys., in press. Metcalfe, T. S., Winget, D. E., & Charbonneau, P. 2001, ApJ, 557, 1021 Metcalfe, T. S., Salaris, M., & Winget, D. E. 2002, ApJ, 573, 803 Montgomery, M. H., Metcalfe, T. S., & Winget, D. E. 2001, ApJ, 548, L53 Nather, R. E., Winget, D. E., Clemens, J. C., Hansen, C. J., & Hine, B. P. 1990, ApJ, 361, 309 Provencal, J. L., Shipman, H. L., Thejll, P., & Vennes, S. ;. 2000, ApJ, 542, 1041 Robinson, E. L., Kepler, S. O., & Nather, R. E. 1982, ApJ, 259, 219 Scargle, J. D. 1982, ApJ, 263, 835 Schwarzenberg-Czerny, A. 1991, MNRAS, 253, 198 Schwarzenberg-Czerny, A. 1999, ApJ, 516, 315 Shipman, H. L., Provencal, J., Riddle, R., Vuckovic, M. 2002, AAS, 200, 7206. Standish, E. M. 1998, A&A, 336, 381 Thompson, S.E., Clemens, J. C., van Kerkwijk, M. H., & Koester, D. 2003, ApJ, submitted. van Kerkwijk, M. H., Clemens, J. C., & Wu, Y. 2000, MNRAS, 314, 209 Vuille, F. et al. 2000, MNRAS, 314, 689 Winget, D. E. et al. 1994, ApJ, 430, 839 Winget, D. E., Robinson, E. L., Nather, R. E., Kepler, S. O., & O’Donoghue, D. 1985, ApJ, 292, 606 Wolff, B., Koester, D., Montgomery, M. H., & Winget, D. E. 2002, A&A, 388, 320 Wu, Y. 2001, MNRAS, 323, 248 Wu, Y., & Goldreich, P. 2001, ApJ, 546, 469 Kepler et al.: The Everchanging Pulsating White Dwarf GD358 17

Table 11. Linear Combination of Peaks in 2000

k Period Amp. fobs Combination fcomb ∆f = fobs − fcomb (s) (mma) (µHz) (µHz) (µHz) 8096.02 3.518 123.52 15 − 17 123.49 0.02 ” ” ” ” 16 − ` = 2 123.38 0.14 ” ” ” ” 17 − 19 123.61 −0.09 6078.39 2.169 164.52 15 − 18 164.68 −0.17 6032.23 1.340 165.78 15 − 18f 165.65 0.12 5669.64 1.152 176.38 1765.91 1.738 566.28 11 − 17 566.27 0.01 1539.52 1.009 649.56 1450.02 2.140 689.64 11 − 19 689.88 −0.23 1369.15 3.584 730.38 10 − 17 730.40 −0.02 1289.88 0.881 775.26 90 − 16 775.23 0.03 1166.29 2.876 857.42 90 − 17 857.42 0.00 1064.99 3.162 938.97 8+ − 15 939.02 −0.04 1056.84 1.121 946.22 8− − 15 946.17 0.04 959.37 0.977 1042.35 90 − 20 1043.07 −0.72 941.28 0.895 1062.39 8+ − 17 1062.51 −0.12 900.84 1.401 1110.07 8− − ` = 2a 1110.03 0.05 20 900.13 2.029 1110.95 8− − ` = 2 1110.86 0.10 853.57 1.816 1171.55 19 852.52 2.740 1172.99 ` = 2h? 798.80 3.662 1251.87 ` = 2 − 3.54µHz ` = 2g? 797.63 5.858 1253.72 ` = 2f ? 797.17 5.330 1254.44 ` = 2 796.55 14.870 1255.41 not 1235µHz ` = 2a? 796.02 7.508 1256.24 ` = 2b? 795.73 1.280 1256.71 ` = 2c? 795.36 3.277 1257.29 ` = 2 + 1.88µHz ` = 2d? 794.75 2.433 1258.26 18 ` = 2 + 2.85µHz ` = 2e? 793.88 1.568 1259.63 18 ` = 2 + 4.22µHz 782.89 1.546 1277.31 781.92 1.350 1278.90 17a 771.68 1.221 1295.87 17 771.25 27.940 1296.60 17b 770.80 1.604 1297.36 759.39 1.205 1316.85 725.70 1.286 1377.98 7 − 17b 1378.13 −0.15 16 725.27 5.157 1378.80 15 + 18` = 2 − 17 1378.76 0.03 724.78 2.688 1379.73 709.03 1.185 1410.38 15+ 704.18 29.720 1420.10 15a 702.44 3.003 1423.62 15+3.52µHz 690.99 1.123 1447.21 12 575.94 1.030 1736.29 11 536.81 0.830 1862.87 10 493.34 1.280 2027.00 9+ 465.01 2.980 2150.49 9+3.54µHz 90 464.25 5.300 2154.03 9− 463.45 2.510 2157.72 9-3.69µHz 447.30 0.968 2235.66 17 + 8+ − 15+ 2235.61 0.08 439.08 0.989 2277.50 15+ + 9 − 17 2277.53 -0.02 18 Kepler et al.: The Everchanging Pulsating White Dwarf GD358

k Period Amp. fobs Combination fcomb ∆f = fobs − fcomb (s) (mma) (µHz) (µHz) (µHz) 8+ 423.89 5.640 2359.11 8− 422.61 5.620 2366.27 8+ − 2 × 3.58µHz 415.34 1.030 2407.65 20 + 17 2407.56 0.09 405.15 1.388 2468.21 19 + 17a 2468.86 −0.65 404.89 1.091 2469.83 19 + 17 2469.60 0.24 398.57 1.249 2508.96 2 × ` = 2f 2508.89 0.08 398.29 1.423 2510.74 2 × ` = 2 2510.83 −0.09 398.15 2.052 2511.65 2 × ` = 2a 2512.49 −0.84 392.10 1.670 2550.34 ` = 2g + 17 2550.32 0.01 392.00 1.737 2551.04 ` = 2g + 17b 2551.08 −0.04 ”””” ` = 2f + 17 2551.05 0.00 ”””” ` = 2 + 17a 2551.28 −0.24 391.85 3.703 2552.02 19 + 16 2551.79 0.23 ”””” ` = 2f + 17a 2551.80 0.22 ”””” ` = 2 + 17 2552.02 0.00 ”””” ` = 2a + 17a 2552.11 −0.09 391.72 2.007 2552.83 ` = 2 + 17b 2552.77 0.05 ”””” ` = 2a + 17 2552.85 −0.02 ”””” ` = 2b + 17a 2552.58 0.24 ”””” ` = 2c + 17 2553.32 −0.49 ”””” ` = 2c + 17a 2553.16 −0.34 391.56 1.159 2553.88 ` = 2a + 17b 2553.60 0.28 ”””” ` = 2b + 17b 2554.07 −0.19 ”””” ` = 2c + 17 2553.90 −0.02 ”””” ` = 2c + 17b 2554.65 −0.77 ”””” ` = 2e + 17a 2554.13 −0.25 385.62 7.759 2593.21 2 × 17 2593.21 0.00 379.48 0.928 2635.16 ` = 2 + 16 2634.21 0.95 ”””” ` = 2a + 16 2635.04 0.12 ”””” ` = 2b + 16 2635.51 −0.35 374.24 1.576 2672.05 19a + 15 2671.97 0.08 374.00 3.459 2673.81 ` = 2g + 15 2673.82 −0.01 373.89 3.498 2674.56 ` = 2f + 15 2674.54 0.02 ” ” ” ” 17a + 16 2674.67 −0.10 7 373.76 8.430 2675.49 17 + 16 2675.40 0.09 373.64 4.254 2676.40 ` = 2a + 15 2676.34 0.06 ”””” ` = 2b + 15 2676.81 −0.41 ” ” ” ” 17b + 16 2676.15 0.25 373.50 1.827 2677.38 ` = 2g + 15a 2677.34 0.04 ”””” ` = 2c + 15 2677.39 −0.01 373.38 1.193 2678.22 ` = 2g + 15a 2678.06 0.16 ”””” ` = 2e + 15 2678.35 −0.13 373.16 0.870 2679.78 ` = 2 + 15a 2679.03 0.75 ”””” ` = 2a + 15a 2679.86 −0.08 ”””” ` = 2b + 15a 2680.33 −0.55 372.83 0.932 2682.18 15a + ` = 2d 2681.87 0.30 368.09 5.653 2716.71 17 + 15 2716.70 0.01 367.62 0.840 2720.17 17a + 15a 2719.49 0.68 ” ” ” ” 17 + 15a 2720.22 −0.05 357.29 1.731 2798.86 16 + 15 2798.89 −0.03 357.16 0.877 2799.88 16a + 15 2799.83 0.05 Kepler et al.: The Everchanging Pulsating White Dwarf GD358 19 k Period Amp. fobs Combination fcomb ∆f = fobs − fcomb (s) (mma) (µHz) (µHz) (µHz) 352.09 4.260 2840.20 2 × 15 2840.19 0.01 351.66 1.054 2843.68 15 + 15a 2843.71 −0.03 293.60 0.992 3405.99 ` = 2 + 9+ 3405.91 0.09 ”””” ` = 2a + 9+ 3406.74 −0.74 ” ” ” ” 16 + 10 3405.79 0.20 289.80 1.929 3450.64 17a + 90 3449.90 0.74 ” ” ” ” 17 + 90 3450.63 0.01 ” ” ” ” 15a + 10 3450.62 0.02 279.78 1.393 3574.19 15 + 90 3574.12 0.07 ” ” ” ” 15a + 9+ 3574.11 0.08 279.50 1.118 3577.77 15 + 9− 3577.82 −0.05 ” ” ” ” 15a + 90 3577.65 0.13 276.66 1.550 3614.54 ` = 2 + 8+ 3614.53 0.01 276.10 0.960 3621.85 ` = 2 + 8− 3621.68 0.17 273.54 1.284 3655.77 17 + 8+ 3655.72 0.05 273.01 1.382 3662.91 17 + 8− 3662.87 0.03 264.60 1.523 3779.25 15 + 8+ 3779.21 0.04 264.11 3.649 3786.37 15 + 8− 3786.37 0.00 ” ” ” ” 20 + 7 3786.44 −0.07 259.84 1.569 3848.54 19 + 7 3848.48 0.06 259.78 0.885 3849.36 2 × 17 + ` = 2 3848.61 0.75 257.08 3.204 3889.80 3 × 17 3889.81 −0.02 254.51 1.744 3929.10 ` = 2g + 7 3929.21 −0.11 ”””” ` = 2 + ` = 2g + 15 3929.22 −0.12 ” ” ” ” 15 + 2 × ` = 2f 3929.06 0.04 254.42 1.140 3930.46 ` = 2f + 7 3929.93 0.53 ” ” ” ” 16 + ` = 2 + 17a 3929.84 0.62 ” ” ” ” 15 + 2 × ` = 2 3930.83 −0.37 254.34 2.986 3931.77 2 × 17 + ` = 2 + 16 3931.76 0.00 251.82 1.015 3971.17 17 + ` = 2g + 15 3970.41 0.75 ” ” ” ” 15 + ` = 2f + 17 3970.44 0.73 ” ” ” ” 15 + ` = 2 + 17a 3971.14 0.03 251.76 2.242 3972.03 15 + 17 + ` = 2 3972.12 −0.08 ” ” ” ” 16 + 2 × 17 3972.01 0.03 ” ” ” ” 17 + 7 3972.09 −0.06 251.70 1.395 3972.93 17 + ` = 2a + 15 3973.01 −0.07 ” ” ” ” 15 + 17a + 16 3972.92 0.01 249.17 1.822 4013.33 2 × 17 + 2 × 15 4013.31 0.02 246.59 0.948 4055.26 15 + 16 + ` = 2 4055.26 0.00 244.27 1.462 4093.91 15 + 2 × ` = 2 4093.90 0.01 244.22 1.576 4094.67 ` = 2g + 2 × 15 4094.64 0.03 ” ” ” ” 17a + 16 + 15 4094.73 −0.06 ” ” ” ” 2 × 15 + ` = 2f 4094.66 0.02 244.16 3.487 4095.59 ` = 2 + 2 × 15 4095.61 −0.02 ” ” ” ” 17 + 16 + 15 4095.47 0.12 ” ” ” ” 15 + 7 4095.58 0.01 244.11 1.722 4096.45 17 + 15 + 16a 4096.48 −0.03 ” ” ” ” 16 + 17a + 16 4096.50 −0.05 210.65 1.196 4747.23 2 × 17 + 90 4747.24 −0.01 198.63 1.272 5034.61 15 + ` = 2 + 8+ 5034.64 −0.03 198.34 1.451 5041.82 15 + ` = 2 + 8− 5041.95 −0.14 20 Kepler et al.: The Everchanging Pulsating White Dwarf GD358

k Period Amp. fobs Combination fcomb ∆f = fobs − fcomb (s) (mma) (µHz) (µHz) (µHz) 196.74 0.992 5082.98 17 + 15 + 8− 5082.97 0.00 192.81 1.319 5186.46 4 × 17 5186.42 0.04 ” ” ” ” 3 × ` = 2 + 15 5185.88 0.59 ” ” ” ” 2 × ` = 2 + 16 + 17 5187.18 −0.72 192.07 1.641 5206.47 2 × 15 + 8− 5206.47 0.00 189.81 1.171 5268.55 2 × 17 + 7 5267.77 0.78 ” ” ” ” 2 × 17 + 15 + ` = 2 5268.64 −0.09 186.94 1.153 5349.20 3 × ` = 2 + 15 5349.33 −0.12 186.85 2.004 5351.84 2 × 7 5350.97 0.87 ” ” ” ” 15 + 16 + 17 + ` = 2 5351.86 −0.02 185.46 0.842 5392.02 2 × 15 + 17 + ` = 2 5392.13 −0.11 181.30 0.609 5515.65 3 × 15 + ` = 2 5515.69 −0.04 165.46 0.533 6043.85 15 + 17 + ` = 2 + 90 6043.83 0.01 162.72 0.584 6145.42 8+ + 8− + 15 6145.48 −0.07 158.96 0.554 6290.83 2 × ` = 2 + 15 + 8+ 6290.02 0.81 ” ” ” ” 8+ + 16 + 17 + ` = 2 6290.88 −0.05 154.76 0.958 6461.78 3 × 90 6462.08 −0.30 ”””” ` = 2 + 2 × 15 + 8− 6461.88 −0.10 ” ” ” ” 16 + 15 + 17 + 8− 6461.77 0.01 147.67 0.761 6771.90 2 × ` = 2 + 3 × 15 6771.06 0.84 ” ” ” ” 2 × ` = 2 + 15 + 16 + 17 6771.94 −0.04 129.69 0.438 7710.99 2 × 15 + 8+ + 8− 7710.93 0.06 ” ” ” ” 8+ + 15 + 16 + 17 + ` = 2 7710.95 0.04 ” ” ” ” 7 + 8+ + 15 + ` = 2 7710.10 0.89 129.61 0.410 7715.61 8− + 3 × ` = 2 + 15 7715.47 0.14 126.87 0.502 7881.92 3 × 15 + ` = 2 + 8+ 7881.87 0.05 125.87 0.410 7944.98 2 × 15 + 2 × ` = 2 + 2 × 17 7944.07 0.91 124.56 0.401 8028.10 3 × ` = 2 + 3 × 15 8027.31 0.78 ” ” ” ” 3 × ` = 2 + 15 + 16 + 17 8027.31 0.78 Kepler et al.: The Everchanging Pulsating White Dwarf GD358 21

FT of GD358 data 2000

Fig. 1. Fourier transform of the 2000 data set. The main power is concentrated in the region between 1000 µHz and 2500 µHz. The marks on top of the graph are the asymptotic equally spaced periods prediction, and the numbers represent the radial order k value, with Winget et al. (1994) identification. The vertical scale on each panel are adjusted to accommodate the large range of amplitudes shown, and the noise level which decreases from 0.29 mma up to 3000 µHz to 0.19 mma upwards. 22 Kepler et al.: The Everchanging Pulsating White Dwarf GD358

Fourier Transform of GD358 data

Fig. 2. Fourier transform of the GD 358 data, year by year. Kepler et al.: The Everchanging Pulsating White Dwarf GD358 23

Fig. 3. The amplitude modulation of GD 358 423 s mode observed in the optical in August, 1996. The timescale of this change is surprisingly short. We have never observed such a fast amplitude modulation in any of the pulsating white dwarf stars. 24 Kepler et al.: The Everchanging Pulsating White Dwarf GD358

Fig. 4. First half of the ground-based optical lightcurve of GD 358 in August 1996. Each panel is one day long. Kepler et al.: The Everchanging Pulsating White Dwarf GD358 25

Fig. 5. Fourier transforms of GD 358 observed one day apart. The top panel shows the Fourier transform of the data taken on the 1st day of the 3-site campaign (suh–55: taken in Poland with start time at 23:28:00 UT on August 10), and the bottom panel shows the data taken about one day later from McDonald (an–0034: taken with start time at 2:48:20 UT on August 12). The observed power has shifted completely and dramatically, both in frequency and amplitude. 26 Kepler et al.: The Everchanging Pulsating White Dwarf GD358

Fig. 6. Fourier transform of an-0034 before (dotted line) and after (solid line) it was prewhitened by the 423 s mode. After prewhitening, there is little significant power left. The lightcurve was dominated by one mode, a possible explanation for why the lightcurve looked so linear (sinusoidal) in Fig. 4. Kepler et al.: The Everchanging Pulsating White Dwarf GD358 27

Fig. 7. GD 358 Fourier transform at four different times along with their spectral windows. The 1994 and 1997 Fourier transforms look similar (within the observed frequency resolution, that is). The September 1996 data look similar as well to these two data sets, but the highest amplitude modes have shorter frequencies (longer period). Obviously, the August 1996 Fourier transform looks very different from the other Fourier transforms. 28 Kepler et al.: The Everchanging Pulsating White Dwarf GD358

Fig. 8. GD 358 lightcurves over time. The shape of the lightcurve was sinusoidal when the amplitude was highest. The 1994 and September 1996 data exhibit similar pulse shapes and their corresponding power spectra also look similar (Fig. 7). Kepler et al.: The Everchanging Pulsating White Dwarf GD358 29

GD358 1990 and 2000 FTs

15

10

5

0

1200 1220 1240 1260 1280

Fig. 9. Peaks around k=18 in the 1990 (solid line) and 2000 (dashed line) transforms 30 Kepler et al.: The Everchanging Pulsating White Dwarf GD358

Peaks present in GD358 data

Fig. 10. Results of pre-whitening for the 1990, 1994, August 1996, and 2000 data sets Kepler et al.: The Everchanging Pulsating White Dwarf GD358 31

GD358 1990 and 2000 FTs 10

8

6

4

2

0

1320 1340 1360 1380 1400

Fig. 11. Peaks around k=16 in the 1990 (solid line) and 2000 (dashed line) transforms 32 Kepler et al.: The Everchanging Pulsating White Dwarf GD358

422.7 8-

422.6

422.5

422.4 1985 1990 1995 2000 2005

423.4 8

423.3

423.2

423.1

1985 1990 1995 2000 2005

424.1 8+ 424

423.9

423.8

1985 1990 1995 2000 2005 Year

Fig. 12. Search for P˙ : The periods of the m subcomponents of even the most stable mode, k = 8, change significantly, from year to year. The same behavior is detected for all pulsations. Kepler et al.: The Everchanging Pulsating White Dwarf GD358 33

GD358 1990 FT

2

1

0

1900 1950 2000 2050 2100

Fig. 13. Peaks around k=10 in the 1990 (solid line) and 2000 (dashed line) transforms 34 Kepler et al.: The Everchanging Pulsating White Dwarf GD358

FT of GD358 data 2000

Fig. 14. Pre-whitened peaks in the 2000 transform